Meta's AI layoff lawsuit exposes discrimination risks Canadian HR leaders can’t afford to ignore, says lawyer
A lawsuit filed in California on July 14, 2026, may have originated in the United States, but the legal question at its centre is one that employment lawyers in Canada are already fielding. Twenty-six Meta Platforms employees have sued the tech giant, alleging that artificial intelligence (AI) systems used to select workers for redundancy disproportionately targeted those on medical, parental, and family leave. The plaintiffs are among the 8,000 employees – approximately 10 per cent of Meta's global workforce – that the company announced for layoffs in May 2026.
The case isn’t a distant curiosity for Canadian organizations, however. Given the rapid pace of AI-enabled transformation in this country, it’s likely to be a preview of things to come on this side of the border, says Aleksandra Pressey, an employment lawyer at Williams HR Law in the Greater Toronto Area.
“A lot of these large organizations that are currently engaged in significant workforce reductions have Canadian offices as well, and other organizations that might be smaller that may be leaning on some of these AI tools face similar pitfalls,” says Pressey. “And certainly on the employee side, we are seeing a lot of AI-generated legal claims, so the existence of AI as a tool that employees and former employees can use as well is going to lead to more claims against employers for having potentially used AI in these decisions.”
Allegations of discriminatory layoff decisions by AI
The lawsuit, filed in the US District Court for the Northern District of California, claims Meta relied on internal AI systems, keystroke and activity-monitoring data, AI token-usage dashboards, and algorithmically generated performance rankings to determine who would lose their jobs. The problem, according to the filing, was structural: those scores, the complaint states, "by design, cannot be accumulated by an employee who is on protected medical or family leave, or whose output is reduced by a disability." The company, the lawsuit argues, "did not pause the system for the individualized, leave- and accommodation-neutral review that the law requires,” reported the Associated Press.
Of the 26 anonymous plaintiffs, roughly half had taken caregiving or pregnancy-related leave. Eight are women who had taken maternity or pregnancy leave; four are men who had taken parental leave, according to the Associated Press. Their terminations were scheduled to begin July 22.
Meta has denied the allegations in full. A company spokesperson said in a statement to media that the claims "lack merit and are not based on facts," adding that "workforce management and organizational decisions were and are made by people, not AI."
Intent is not the test
Under Canadian human rights legislation, the threshold for a discrimination finding is not whether an employer meant harm, says Pressey.
The scenario playing out at Meta illustrates how AI tools can generate that liability without anyone recognizing it. If an organization uses AI to assess its workforce and identifies a role as redundant – in part because the person in it has been on a protected leave and their measurable output has dropped – the process may have crossed a legal line, says Pressey. The same risk arises when a termination decision is made and only afterwards does the employer learn that the affected employee is pregnant, on disability leave, or requesting an accommodation. Without contemporaneous documentation proving the decision predated that disclosure, the exposure is significant, according to Pressey.
That defence, Pressey notes, depends entirely on what was written down and when. A timestamped note, a message to a colleague, or any record demonstrating that a performance concern or restructuring rationale existed independently of a human rights matter is often the difference between a claim that resolves quickly and one that becomes protracted and reputationally costly.
“The reality is just because there is a protected ground at play doesn't mean you can't reasonably and in a non-discriminatory manner make a decision to terminate somebody's employment,” says Pressey. "You can very rarely stop somebody making a claim, so what you want to be able to do is defend it."
Three questions every HR leader should answer
When an HR executive tells Pressey their organization uses algorithmic productivity scoring to inform workforce decisions, she has a standard line of inquiry. "I would ask them to explain what metrics they're looking at specifically what mechanism they have for verifying those findings and checking them," she says. “And then also, if they do keep additional lists or data about any employees, who might have some type of accommodation or if things like legitimate job-protected workplace absences have been included in that consideration.”
The concern is that productivity metrics frequently fail to capture what they purport to measure. If output is assessed by keystrokes logged or AI tokens used, high-performing employees who work efficiently will often appear underproductive, says Pressey. "Some employees produce more output in less time, so are you accounting for that?" she says. "It's the how and the why that I would be interested in."
The more fundamental principle, she argues, comes down to accountability. "AI is helping to give you the information you need to make a decision," says Pressey. "AI shouldn't be making the decision for you."
The shadow AI risk
One liability exposure that many Canadian organizations haven’t yet fully considered extends beyond tools that appear in formal policy. When leadership instructs managers to reduce their teams by a given percentage, those managers make individual calls – and the tools they use to make them aren’t always disclosed or monitored, says Pressey.
"Do you know if one of those managers is on the side using their personal ChatGPT subscription to help them make that call?" she says. "If you don't, then there’s liability there – and I would hazard a guess that not everybody can answer that question."
During the big push towards AI integration and transformation, organizations may still lack proper AI understanding and oversight, adds Pressey, so if a machine is analyzing metrics for which someone with an accommodation need has a lower output and doesn’t recognize that, the liability risk increases.
“If you aren't sure of the input, then the output is problematic,” she says. “And so you very well could, as an organization trying to automate some of those things, run into these issues if you're not vetting these scenarios.”
Canadian AI termination case likely coming
Pressey reiterates the likelihood that Canada will see its own version of the Meta case. "I think unless things change course significantly that's going to be coming," she says. “We also don't know what decisions have already been made and what things may be in the works.”
She points out that limitation periods in Ontario, for example, allow wrongful dismissal claims up to two years after termination and human rights claims up to one year, meaning decisions made in recent months by companies with Canadian operations may not yet have surfaced as formal complaints.
"The danger here is complacency," says Pressey. "We put it into AI, so the decision must be correct. If you're staying on your toes and monitoring those outputs and the inputs, then there’s less risk of this kind of situation occurring."
